Spatio-temporal Modeling of Residential Sales Data

Equity in the personal residence is the largest single asset in the investment portfolio of most households. Lenders also have keen interest in residential property values. Estimating the market value of an individual house through appraisal is quite challenging. Regression analysis based upon hedonic pricing theory using individual house characteristics ooers an attractive alternative. The availability of residential transaction data bases makes such analysis feasible but coincides with unprecedented volatility in housing prices in many markets. Homeowners and lenders need to know the value of an individual house in a particular location at a speciied point in time. This paper focuses on the location, time and spatio-temporal components associated with suitably aggregated data to improve prediction of individual asset values. Such eeects are introduced in the context of hierarchical models which we nd more natural than attempting to model covariance structure. Indeed, our cross-sectional data base, a sample of 7936 transactions for 49 subdivisions over a 10 year period in Baton Rouge, Louisiana, precludes covariance modeling. A wide range of models arises, each tted using sampling-based methods since likelihood based tting may not be possible. Choosing amongst an array of nonnested models is carried out using a posterior predictive criterion. In addition, one year of data is held out for model validation. A thorough analysis of the data incorporating all of the aforementioned issues is presented. .